Bidirectional adaptive compression

A new dynamic Huffman encoding has been proposed recently (Shmuel et al., 2021), which instead of basing itself on the information gathered from the already processed portion of the file, as traditional adaptive codings do, uses rather the information that is still to come. The current work extends this idea to bidirectional adaptive compression, taking both past and future into account, and not only performs at least as good as static Huffman, but also provably improves on the future-only based variant. We extend the technique to arithmetic coding and give both theoretical and empirical results that support the enhancement of the new compression algorithm.